The LinkedIn MCP Server for AI Agents
Point Claude, ChatGPT, or any AI agent at LinkedIn: find people, send invites and messages, and pull enriched profiles, on accounts that stay safe by design.
A LinkedIn MCP server is a typed set of tools an AI agent calls to act on LinkedIn: search, connect, message, and enrich. The agent-native, account-based LinkedIn API is GTM API: connect Claude, Cursor, or ChatGPT to one endpoint and your agent runs real LinkedIn outreach through 160+ typed tools, with the account safety enforced server-side. It’s the same infrastructure GTM API reports running 20,000+ LinkedIn accounts at under a 1% ban rate.
Drafting the message is the easy half. An MCP server is the layer that lets the agent do the work itself (send the invite, reply in the thread, enrich the lead) with no browser extension and no scraping. The rest of this page covers what an MCP server is, how to connect it to Claude, what the agent can do, and why it doesn’t get your account banned.
On this page
What is a LinkedIn MCP server?
A LinkedIn MCP server is a program that exposes LinkedIn actions (search, connect, message, enrich) as Model Context Protocol tools that an AI agent can call. Instead of a person clicking through a dashboard, the agent reads the tool list and runs the action itself. A managed one is GTM API: hosted and typed, wired into the safety stack that keeps accounts alive.
MCP is the open standard introduced by Anthropic in late 2024 and since adopted across Claude, OpenAI’s ChatGPT, Cursor, and the wider agent ecosystem. People describe it as “a USB-C port for AI”: one connector that lets any model plug into any tool. A LinkedIn MCP server applies that standard to a single channel, LinkedIn.
At its core, GTM API is MCP-first: every capability is defined once as a typed schema, which generates the MCP tools, a REST API, and webhooks from the same contract. A Claude or Cursor agent and a custom backend therefore act on LinkedIn through exactly the same validated operations, with no drift between them and no glue code to maintain.
How to connect LinkedIn to Claude
Connect LinkedIn to Claude in two steps: open Settings → Connectors → “Add custom connector”, paste https://mcp.gtm-api.com/mcp, and approve the consent screen that opens in your browser. There is no key to paste. The server advertises its authorization server and your client registers itself, so consent is the whole credential. Claude Code takes one command, claude mcp add --transport http GTM API https://mcp.gtm-api.com/mcp, then /mcp in a session to authenticate. Per-client setup, the config-file trap and troubleshooting are on Connect LinkedIn to Claude.
Prefer a local CLI?
Clients that only speak stdio, and headless agents with no browser for the consent screen, launch the same server locally through the npm CLI. Create an API key in the app, then it is one line, no clone, no config file:
GTM_API_KEY=gtm_live_xxx npx -y @gtm-api/linkedin-mcp
Then prompt in plain English
Once the connector reports connected, just ask. The agent sees the LinkedIn tools and picks the right ones:
“Every morning, accept new connection invitations from founders, reply with a short welcome, and add anyone hiring SDRs to a ‘warm’ list.”
Under the hood the agent calls get_my_latest_linkedin_connection_invitations, accept_linkedin_connection_invitation, and send_linkedin_message. Every outward action runs a preview → confirm step and a server-side daily-limit check before it ever touches LinkedIn.
Can ChatGPT or an AI agent send LinkedIn messages?
Yes. Through a LinkedIn MCP server, ChatGPT, Claude, or a custom agent can actually send LinkedIn messages, connection requests, and InMail, not just draft them. For that, GTM API exposes send_linkedin_message, send_linkedin_connection_request, and send_linkedin_inmail as tools. The agent calls them and the message goes out from your account.
Teaching a model to write the message still leaves you pasting it in by hand. An MCP server closes the loop. The agent sends it, reads the reply with get_my_latest_linkedin_messages, and can hand a thread back to a human by marking it unread when judgment is needed, so a person still owns the calls that matter. Plenty of tools can “send” through MCP now, HeyReach and Salesforge among them, so sending is no longer the differentiator. What separates the tools is where the send fires from: an owned, warmed account, behind a preview → confirm step and a server-side daily-limit check.
What can the agent do?
A GTM API agent can run the full LinkedIn outreach loop: find people, send and accept connections, message across every LinkedIn surface, track and engage with posts, enrich profiles and companies, and read its own account health. That’s 160+ typed tools across 11 LinkedIn toolsets, all behind one key.
| Toolset | What your agent can do |
|---|---|
| Messaging | Send DMs, voice notes, InMail and Sales Navigator messages. Sync and search the whole inbox. Read reply metrics |
| Network | Send & withdraw connection requests, accept/ignore invitations, list connections, followers and their growth |
| Content | Track posts and their metrics, comment, like/react, pull the engagers and commenters on any post |
| Enrichment | Lite & full profile, experience, skills, education, posts, plus company data, returned as typed JSON |
| Scraping / search | People, company and post search by URL or params. Similar profiles, employees, decision-makers, post engagers |
| Accounts & pacing | Connect accounts, set smart daily limits, read the self-profile, reset a stale sync |
| Account monitor | Health snapshots, benchmarks, quota-hit log, block log, activity log, sync runs (read-only audit) |
| Browsers | Anti-detect cloud browsers and dedicated proxies with connectivity & location checks |
| Data plane | Saved searches (create / run), the credit ledger, and stored result rows |
| Platform | Webhooks + delivery logs, so your agent reacts to LinkedIn events in real time |
LinkedIn is the live channel today. Email (Gmail, Outlook, IMAP), messengers (WhatsApp, Telegram, Instagram DMs) and calendars (Google, Microsoft) are on the roadmap on the same typed contract, so an agent built against LinkedIn today will pick up those channels as they land.
Is it safe for my LinkedIn account?
Yes. The stack is built around account safety: GTM API runs each account in its own anti-detect cloud browser with a dedicated residential proxy, warms new accounts up, and enforces health-aware daily limits server-side, before any action fires. On this stack, GTM API reports 20,000+ LinkedIn accounts at under a 1% ban rate, self-reported. MirrorProfiles‘ co-founder calls the stack behind it “the safest and most reliable option we’ve tested”.
The safety comes from four things working together: a single dedicated session per account (no shared browser fingerprints), a residential proxy pinned to the account’s location, a warm-up curve on new accounts, and a limit check that runs on every send_* call so an over-eager agent can’t burn an account. Outward actions also require preview → confirm, and every action lands in an activity and quota-hit log you can audit. Compare that to a browser extension running in your own logged-in Chrome, or a scraper hammering pages with your cookie: those are the exact patterns that trigger restrictions.
LinkedIn MCP server vs open-source scrapers
A managed LinkedIn MCP server runs on owned, warmed accounts behind anti-detect sessions, dedicated residential proxies and server-side limits, so it holds up at volume. An open-source MCP server drives a real browser session with your own cookie and no safety layer. That holds up for a weekend demo. Point real outreach volume at it and the missing safety layer becomes the problem. The popular open-source options, like stickerdaniel/linkedin-mcp-server on GitHub (1.9k★), Apify‘s LinkedIn actors, and the servers listed on mcpservers.org and Docker’s MCP catalog, prove the pattern but ship no account safety. (Our own server is in the same public places: gtm-api/linkedin-mcp on GitHub, @gtm-api/linkedin-mcp on npm, and a Docker image.) Other managed servers take the same “safe managed” angle, most visibly ConnectSafely.ai. What GTM API brings that the rest of the list doesn’t is at-scale numbers it publishes: a self-reported under-1% ban rate across 20,000+ owned accounts, plus a typed MCP + REST contract to build against. And unlike scraper-APIs that were sued and shut down (Proxycurl, 2025), owned accounts and a published ban-rate don’t disappear. Open-source is the right place to experiment, and production outreach belongs on a managed API.
| Dimension | gtm-api.com (managed) | Open-source MCP | Scraper / Apify MCP |
|---|---|---|---|
| Setup | One key, hosted | Self-host + your cookie | Actor + your cookie |
| Safety layer | Anti-detect + proxy + limits | None | None |
| Daily-limit enforcement | Server-side, per account | You build it | No |
| Account model | Owned, warmed, isolated | Your live session | Your live session |
| Coverage | 160+ tools, send + read + enrich | Reads, some sends | Reads / scrape only |
| Ban risk at volume | <1% (20k accts) | High | High |
| Support & uptime | Managed SLA | Community | Best-effort |
Open-source servers currently win the “linkedin mcp” search result but not the production workload. They prove the pattern, then teams move to a managed stack once an account gets restricted. Among managed options, a published ban-rate at scale is the differentiator.
What does it cost?
At volume, GTM API is from $10 per connected account per month, with unlimited API calls and no per-action fees. The forever free plan needs no card and there’s no minimum. Scraping and enrichment are live, with pricing not yet published. Billing is per connected account, so the number of messages your agent sends doesn’t change the price.
Full breakdown, including how per-account pricing compares to the per-call metering most data APIs use, is on the LinkedIn API pricing page.
Frequently Asked Questions
Is there a LinkedIn MCP server?
Yes: GTM API is a managed LinkedIn MCP server, a hosted set of Model Context Protocol tools an AI agent calls to search, connect, message and enrich on LinkedIn. There are also open-source LinkedIn MCP servers on GitHub, but they scrape with your cookie and have no safety layer, so they are better for demos than production.
Which AI clients support the GTM API LinkedIn MCP?
Claude (Desktop and API), Cursor, ChatGPT, LangChain and n8n, any MCP-compatible client. For everything else, the same actions are available over REST and webhooks, because GTM API is built from one typed schema that generates MCP, REST and webhooks together.
Do I need to write code to use it?
No. In an MCP client like Claude or Cursor you connect the server once, then prompt in plain English. The agent discovers the LinkedIn tools and runs them. Developers who want to build a product on top can call the same actions over the REST API instead.
Can ChatGPT or an AI agent actually send LinkedIn messages?
Yes. Through the MCP server the agent calls real tools, send_linkedin_message, send_linkedin_connection_request, send_linkedin_inmail, so the message is sent from your connected account, not just drafted. Every send runs a preview-then-confirm step and a server-side daily-limit check first.
Is a LinkedIn MCP server safe for my account?
It depends entirely on the server: GTM API runs each account in its own anti-detect cloud browser with a dedicated residential proxy, warms new accounts up, and enforces health-aware daily limits before any action. On that stack, gtm-api.com reports 20,000+ LinkedIn accounts at under a 1% ban rate. Cookie-based open-source servers offer none of that.
Can I connect more than one LinkedIn account?
Yes, that is the design. Connect as many accounts as you like. Each runs in an isolated cloud browser with its own proxy and its own pacing. You pay per connected account per month, and your agent can act across all of them through one key.
Is there a free trial?
There’s no trial because the free plan doesn’t expire: Sandbox, the forever free plan, covers one connected account with full API and MCP access, no credit card required and no minimum commitment. Pricing is from $10 per connected account per month at volume, with unlimited API calls.
Is LinkedIn automation against LinkedIn’s terms?
Third-party automation is a grey area and runs against LinkedIn’s User Agreement, so it always carries some risk, and no tool can honestly promise otherwise. That’s exactly why a per-account safety layer matters: GTM API runs each account in an isolated cloud browser on its own residential proxy with server-side limits. On that setup, gtm-api.com reports 20,000+ accounts at under a 1% ban rate.
Read next on this topic
Sources & Further Reading
- Anthropic / Model Context Protocol, modelcontextprotocol.io (the open standard)
- gtm-api.com, product & live LinkedIn toolsets
- MirrorProfiles, LinkedIn automation guide (vendor guide, co-founder testimonial)
- Example open-source server, stickerdaniel/linkedin-mcp-server (GitHub) · directories: mcpservers.org, MCP Market · our listings: Glama, Smithery
- Related: The LinkedIn API · Safe LinkedIn automation · LinkedIn API pricing · LinkedIn API access · Connect LinkedIn to Claude · LinkedIn integrations
Give your agent LinkedIn access.
Sign up, connect your first LinkedIn account, and point Claude, ChatGPT, or Cursor at it in minutes. 160+ typed tools, one contract, the same stack behind the published ban-rate.
Last updated: September 2026 · Tool names and limits verified against the live GTM API at time of publication